An Artificial Intelligence System Based Power Estimation Method for CMOS VLSI Circuits
Govindaraj Vellingiri, Ramesh Jayabalan · 2020
The remarkable reduction in transistor size and the subsequent increase in the number of devices on a single chip in combination with the budding demand for portable devices has increased the power consumption of the chips, which leads to a major challenge in very large scale integration (VLSI) circuit design. During the design phase we need to minimize power dissipation in circuits, which requires accurate estimates of the power dissipated. This will avoid complicated and expensive redesign that might be required due to power constraint violations. A particular strength of artificial intelligence is adaptivity, since it is capable of adjusting to various environmental conditions and changing its behavior accordingly. Higher-level power estimation using artificial intelligence techniques has merit over the traditional power estimation techniques such as reduction in time bound and circuit complexity. Artificial intelligence techniques such as back-propagation neural network (BPNN) and adaptive neuro-fuzzy inference system (ANFIS) are some of the most attractive tools to solve the complex, nonlinear, and time-constraint problem, hence they are well suited for power estimation for CMOS VLSI (complementary metal oxide semiconductor very large-scale integration) circuits. Applying ANFIS for power estimation is a relatively new technique. The major objective of this work is to provide an alternative solution to estimate the power dissipation of CMOS VLSI circuits, using statistical tools such as BPNN and ANFIS, and to overcome the drawback of time complexity through the simulation of complex circuits using traditional power estimator tools. Based on the experimental results, ANFIS gives a better result in terms of testing error that varies from 0% to 0.86% when compared to BPNN and it also has a low root mean square error (RMSE) value of 0.0002075 and a very high coefficient of determination (R) value of 0.99961. Therefore, ANFIS offers an alternative approach to the conventional techniques like SPICE (simulation program with integrated circuit emphasis), which are based on the assumption of predefined empirical equations that depend on arbitrary parameters.